Human gait can be measured using a change in the joint angles value of the lower limb joints during walking. Inverse kinematics (IK) of a human leg refers to calculating joint angle values from different leg positions during locomotion. This paper calculates a real-time IK for a 3-link kinematic leg using a musculoskeletal model in OpenSim. The model provides the analytical solution of IK, which is quick and accurate. But, the model-based approach is very much dependent on the morphology of the bipedal robot. The existing bipedal robots are mainly planned for walking on flat terrain and unconstrained environment. So, these limitations lead to the developing of a generalised polynomial equation for generating walking trajectories. The model-based approaches rely on the degree of similarity between the virtual human model and the actual human body, which does not fully describe the human body. Due to the difference in real and virtual models, the model-based approaches suffer during the optimisation process of the boundary conditions and the mass-inertial parameters, leading to an unstable numerical solution. The learning-based method using long short-term memory (LSTM) model for gait generation is proposed to overcome the limitation of model-based approaches.

In this study, novel modular shape memory alloy wire–based torsional actuators were designed and fabricated. These shape memory alloy–based actuators provide rotational displacements. The mechanical and thermal properties of a single module shape memory alloy torsional actuator were characterized. Next, a modular shape memory alloy torsional actuator was configured by connecting single actuator modules in series. This modular actuator can be used directly as a soft or biologically inspired robot. Finally, the rolling motion and shape transformation of the modular shape memory alloy torsional actuator (soft robot) were demonstrated with a simple open-loop-based control scheme experimentally validating the novel mobility and actuation of the proposed actuator. For a control input, sequentially coordinated square waves of electric current were supplied to the shape memory alloy actuating units.

In this study, we present a novel Shape Memory Alloy (SMA)-actuated robotic hand which can provide a higher level of dexterity and compliance. The mechanical structure and actuation of the robotic hand are designed similar to the musculoskeletal system of a human hand. SMAs are used in a similar manner where muscles and ligaments are applied in a human hand. In our design, the compliance of SMAs are used to manipulate soft, fragile, and/or rigid objects more adaptively and naturally. In particular, SMA actuators can provide a high power-to-weight ratio, which is preferable for a compact and lighter design. Each finger joint in the hand is actuated independently. Thus, the robotic hand can achieve dexterous in-hand manipulation. The grasping and in-hand manipulation experiments verify the robot design using open-loop control and closed-loop control, respectively.

This research advances the field of Human Activity Recognition (HAR) by developing a robust and interpretable deep-learning model using wearable sensor data. We address seven discrete activities through a multimodal fusion architecture that synergistically combines Temporal Convolutional Networks (TCNs), Convolutional Neural Networks (CNNs), and Long Short-Term Memory (LSTMs). Each network type caters to its strength: TCNs for temporal dependencies, CNNs for local features, and LSTMs for sequential information. A dedicated fusion layer seamlessly integrates these features, achieving a remarkable mean accuracy of 98.7% on challenging data. Finally, 5-fold cross-validation is done to validate our results. We find a mean accuracy of 98.7% and a standard deviation of 0.003. In addition, we utilize Local Interpretable Model-Agnostic Explanations (LIME) and Shapley Additive Explanations (SHAP) to offer insights into the model’s decision-making process, thereby improving its transparency and fostering confidence. This study contributes by providing robust and interpretable deep-learning models that can be employed in various applications.